Dynamic Population Distribution Aware Human Trajectory Generation with Diffusion Model

TL;DR

This work tackles generating realistic human trajectories under dynamic population distributions, addressing privacy and data quality challenges. It introduces a diffusion-model framework guided by a dynamic population constraint, aided by a spatial enhancement module that uses LINE-based embeddings to encode geographic proximity. A population-aware noise predictor integrates temporal dynamics and population signals via a Transformer-based cross-attention mechanism, with a joint loss that combines trajectory fidelity and distribution alignment. Extensive experiments on Geolife and MME show clear advantages over state-of-the-art baselines, especially in matching population distributions and origin–destination patterns, highlighting the method’s practical utility for privacy-preserving mobility synthesis and urban planning simulations.

Abstract

Human trajectory data is crucial in urban planning, traffic engineering, and public health. However, directly using real-world trajectory data often faces challenges such as privacy concerns, data acquisition costs, and data quality. A practical solution to these challenges is trajectory generation, a method developed to simulate human mobility behaviors. Existing trajectory generation methods mainly focus on capturing individual movement patterns but often overlook the influence of population distribution on trajectory generation. In reality, dynamic population distribution reflects changes in population density across different regions, significantly impacting individual mobility behavior. Thus, we propose a novel trajectory generation framework based on a diffusion model, which integrates the dynamic population distribution constraints to guide high-fidelity generation outcomes. Specifically, we construct a spatial graph to enhance the spatial correlation of trajectories. Then, we design a dynamic population distribution aware denoising network to capture the spatiotemporal dependencies of human mobility behavior as well as the impact of population distribution in the denoising process. Extensive experiments show that the trajectories generated by our model can resemble real-world trajectories in terms of some critical statistical metrics, outperforming state-of-the-art algorithms by over 54%.

Figures (6)

• Figure 1: Relationship between dynamic population distribution and individual trajectories.
• Figure 2: (Left) The overview of the Trajectory Diffusion Model (TDM). The model is given a noisy trajectory embedding $e^{t}_{1:N}$ of length $N$ in a diffusion step $t$, as well as the diffusion embedding of diffusion step $t$ and population distribution condition $p_{1:N}$. The denoising network predicts the ultimate clean trajectory embedding $\hat{e}^{t}_{1:N}$ in each sampling step. (Right) Sampling TDM. We sample random noise $e^T$ given a population distribution condition $p_{1:N}$, and then iterate from $T$ to 1. TDM predicts the clean sample $\hat{e}^{0}$ at each step $t$ and diffuses it back to $e^{t-1}$.
• Figure 3: Comparative analysis of real vs. generated population distributions across a 64$\times$64 grid urban layout.
• Figure 4: Comparative analysis of real vs. generated population distributions across a 32$\times$32 grid urban layout.
• Figure 5: Comparative analysis of real vs. generated population distributions across a 16$\times$16 grid urban layout.